Sapling MCP Server

The Sapling MCP Server makes Sapling's language APIs available as tools in Claude Desktop, Claude Code, ChatGPT connectors, and other assistants that speak the Model Context Protocol.

There are two ways to connect:

	Hosted server	Local server
Address	https://mcp.sapling.ai/mcp	runs on your machine via npx
Transport	Streamable HTTP	stdio
Requires	An MCP client that supports remote servers	Node.js 18+
Sign-in	Sapling account (OAuth) or an API key header	API key in the client's environment

Use the hosted server unless your client only supports local stdio servers, or you need requests to go to an on-premises Sapling deployment.

Hosted server

Add https://mcp.sapling.ai/mcp as a remote MCP server in your client. On first connection the client opens a Sapling page in your browser where you sign in and choose which of your API keys the connection should use. Approve it, and the client receives a token — no key is ever pasted into the client.

Claude Code:

claude mcp add --transport http sapling https://mcp.sapling.ai/mcp

Then run /mcp in Claude Code and choose Authenticate to complete sign-in.

In Claude Desktop or claude.ai, add a custom connector with the same URL.

Connecting with an API key instead

Clients that let you set request headers can skip the sign-in flow and send a Sapling API key directly:

claude mcp add --transport http sapling https://mcp.sapling.ai/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Requests without a credential are refused with 401.

What a connection can do

An approved connection can call the Sapling tools listed below using the API key you selected. Usage counts against that key's quota and billing, exactly as your own API calls do. The connection cannot read your documents, team data, or account settings, and it cannot create or change API keys.

Access tokens last 24 hours and are renewed automatically by the client for up to 90 days. To disconnect an application, delete or rotate the key it uses in API settings, or remove the server from your MCP client.

Local server

The local server runs through Node.js. When you call a Sapling tool, it sends the requested text to the Sapling API and returns the API's structured response to your assistant.

Before you begin

You need:

1. Node.js 18 or newer, including the npx command.
2. A Sapling API key. Trial keys are available after registration.
3. An MCP client that supports local stdio servers.

Keep the API key in your MCP client's environment configuration. Do not place it in prompts, source code, or a committed configuration file.

Claude Desktop

Add the following entry to your Claude Desktop configuration, replacing YOUR_API_KEY with your Sapling API key:

{
  "mcpServers": {
    "sapling": {
      "command": "npx",
      "args": ["-y", "@saplingai/mcp-server"],
      "env": {
        "SAPLING_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

The configuration file is located at:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop after saving the file.

Claude Code

Run:

claude mcp add sapling --env SAPLING_API_KEY=YOUR_API_KEY \
  -- npx -y @saplingai/mcp-server

Then run claude mcp list to confirm that sapling is configured.

Tool permissions

Claude Code asks for approval the first time Claude calls each Sapling tool. Choose the "don't ask again" option to approve a tool permanently.

To pre-approve tools — or if a tool call is denied instead of prompted, which can happen in auto-accept or non-interactive sessions — run /permissions in Claude Code and add an allow rule:

• mcp__sapling allows every Sapling tool.
• mcp__sapling__sapling_grammar_check allows one tool.

Rule names follow mcp__<server>__<tool>, where the server name is the one you passed to claude mcp add (sapling above). MCP permission rules do not support wildcards; use mcp__sapling to allow all Sapling tools.

You can also add the rules directly to the permissions.allow list in ~/.claude/settings.json (all projects) or .claude/settings.json (one project):

{
  "permissions": {
    "allow": ["mcp__sapling"]
  }
}

Other MCP clients

Use the following local-server settings in any client that accepts a command, arguments, and environment variables:

Setting	Value
Transport	stdio
Command	npx
Arguments	-y, @saplingai/mcp-server
Environment	SAPLING_API_KEY=YOUR_API_KEY

With this configuration your client starts the local stdio process and the process calls https://api.sapling.ai. For a remote server, use https://mcp.sapling.ai/mcp as described in Hosted server.

Configuration

These environment variables configure the local stdio server. The hosted server is configured through its sign-in flow or Authorization header instead.

Environment variable	Required	Default	Description
SAPLING_API_KEY	Yes	—	Sapling API key used for each request
SAPLING_API_BASE_URL	No	https://api.sapling.ai	Alternate Sapling API base URL, such as an on-premises deployment
SAPLING_MCP_TIMEOUT_MS	No	120000	Positive request timeout in milliseconds

Available tools

The two servers share most tools but not all of them, and two tools are spelled differently on each — the hosted names were published first and cannot change without breaking connected clients.

Tool (hosted server)	Purpose	Optional inputs	On the local server
sapling_grammar_check	Grammar, spelling, punctuation, and style suggestions	session_id, lang, variety, auto_apply, medical, advanced_edits	same name
sapling_spellcheck	Spelling-focused suggestions	session_id, lang	sapling_spell_check
sapling_ai_detect	AI-generated content detection (see the AI detector API)	sent_scores, score_string	same name
sapling_rephrase	Paraphrase, tone, and tense transformations	mapping, tone_mapping, tense_mapping, num_results (1–8)	same name
sapling_summarize	Summarize longer text	None	same name
sapling_sentiment	Analyze sentiment	None	same name
sapling_detect_language	Detect a text's language, with runner-up candidates and an optional per-segment breakdown for mixed-language text	top_k (1–10), segments	sapling_language_detect
sapling_detect_pii	Detect PII (emails, phones, SSNs, cards, IPs, IBANs, routing numbers) with exact offsets and optionally return a redacted copy	types, redact	same name
sapling_quality	Score writing quality 1–5, optionally per sentence and with an LLM rubric (clarity, coherence, correctness, concision, quoted issues with suggested rewrites)	sentence_scores, rubric	same name
sapling_seo	SEO analysis and metadata: on-page stats, keyword density, top terms, and suggested title tags, meta descriptions, slug and focus keywords	keywords (up to 10), suggestions, lang	same name
sapling_classify	Zero-shot classification into labels you supply (single- or multi-label), with a score per label and a one-sentence rationale	labels (required, 2–20), multi_label, threshold, context	same name
sapling_extract	Extract fields you name from unstructured text, with a typed value and the quoted source span for each	fields (required, 1–20), context	same name
sapling_tone	Classify emotional tone, overall and per sentence	None	not available
sapling_readability_statistics	Grade level, reading time, and text counts	lang	not available
sapling_autocomplete	Predict how a partial sentence continues	session_id	not available

Every tool requires a non-empty text input, except sapling_autocomplete, which takes query (200 characters maximum) and needs a key with autocomplete enabled. For endpoint-specific limits and response formats, see the API documentation.

Prompts and reference resources

The hosted server also publishes ready-made prompts — proofread, adjust tone, check for AI-generated content, summarize, and readability report — which clients surface as slash commands or quick actions, and reference resources under sapling://reference/ listing the language codes and argument values each tool accepts.

Try it

After restarting or reconnecting your MCP client, try:

• "Check this sentence for grammar: This are a test."
• "Classify this support ticket as billing, technical issue, shipping or other."
• "Detect the language of: Bonjour tout le monde."
• "Pull the invoice number, total and due date out of this email."
• "Rewrite this sentence in a friendlier tone."
• "Suggest a title tag and meta description for this article, targeting grammar checker."
• "Analyze the sentiment of this customer message."
• "Redact any PII from this support transcript before I paste it into the ticket."

Your assistant should ask for confirmation when appropriate and then call a tool whose name begins with sapling_.

Troubleshooting

Missing SAPLING_API_KEY environment variable

Add SAPLING_API_KEY to the server's env configuration and restart the MCP client. Setting it only in a shell that does not launch the client may not pass it to the server.

npx or node is not found

Install Node.js 18 or newer and restart the MCP client so it receives the updated system path.

402 Payment Required

The API key may be expired, inactive, or over its monthly billing limit. Review API access and payment-required responses.

Claude Code denies a Sapling tool call

If a sapling_* tool call is blocked with a permission error rather than an approval prompt, add an allow rule for the tool: run /permissions and allow mcp__sapling (all Sapling tools) or mcp__sapling__<tool_name> (one tool), then retry. See Tool permissions.

The hosted server returns 401

The request carried no credential, or the token expired and the client did not renew it. Reconnect the server in your client to sign in again, or check the Authorization: Bearer header if you configured a key directly.

A tool times out

For unusually long API calls, set SAPLING_MCP_TIMEOUT_MS to a larger positive number. For large documents, follow the chunking guidance on the relevant API page.

Security and data handling

Neither server logs nor persists your API key or submitted text. The hosted server stores no text and no credentials: it forwards your token or key to the Sapling API with the request and keeps nothing afterward. Tool inputs are sent to the configured Sapling API endpoint for processing. Review Data Processing and contact Sapling if you require no-data-retention or an on-premises deployment.

If you set SAPLING_API_BASE_URL, the server sends your API key and submitted text to that endpoint. Point it only at a Sapling endpoint you trust, such as your own on-premises deployment.

Releases are published on npm as @saplingai/mcp-server.